This paper deals with detection and recognition of matrix codes, such as the QR codes, in high-resolution images of real-world scenes. The goal is to provide a detector capable of operation in real time even on high-resolution images (several megapixels). We present an efficient algorithm for detection of possible occurrences of the codes. This algorithm is characterized by a very low false negative rate and a reasonable false alarm rate. The results of our algorithm are to be followed by an accurate detection/recognition algorithm. We propose to use a recent matrix code detection and recognition algorithm based on Hough transform, because it can reuse some information computed by our new pre-detection algorithm and thus a further reduce of computational demands can be achieved. Since there are no publicly available annotated datasets for evaluation of this kind of algorithm, we collected a number of images and annotated them; these images will be made publicly available to allow for a proper comparison. Our algorithm was evaluated on this dataset and the results are reported in the paper.

No abstract

No abstract

A major limitation of contemporary fiduciary markers is that they are either very small (they try to represent a single point in the space) or they must be planar in order to be reasonably detectable. A deformable large-scale marker or marker field that would be efficiently detectable is the objective of this work.We propose a design of such a marker field -the Honeycomb Marker Field. It is composed of symmetric hexagons, whose triplets of modules meet at "Y-junctions". We present an efficient detector of these image features -the Y-junctions. Thanks to the specific appearance of these synthetic image features, the algorithm can be very efficient -it only visits a small fraction of the image pixels in order to detect the Y-junctions reliably. The experiments show that compared to a general feature point detector (FAST was tested), the specialized Y-junctions detector offers better detection reliability.

No abstract